Instrumental Variables. Application and Limitations
|
|
- Adam Singleton
- 5 years ago
- Views:
Transcription
1 ORIGINAL ARTICL Application and Limitations dwin P. Martens,* Wiebe R. Pestman, Anthonius de Boer,* Svetlana V. Belitser,* and Olaf H. Klungel* Abstract: To correct for confounding, the method of instrumental variables (IV) has been proposed. Its use in medical literature is still rather limited because of unfamiliarity or inapplicability. By introducing the method in a nontechnical way, we show that IV in a linear model is quite easy to understand and easy to apply once an appropriate instrumental variable has been identified. We also point out some limitations of the IV estimator when the instrumental variable is only weakly correlated with the exposure. The IV estimator will be imprecise (large standard error), biased when sample size is small, and biased in large samples when one of the assumptions is only slightly violated. For these reasons, it is advised to use an IV that is strongly correlated with exposure. However, we further show that under the assumptions required for the validity of the method, this correlation between IV and exposure is limited. Its maximum is low when confounding is strong, such as in case of confounding by indication. Finally, we show that in a study in which strong confounding is to be expected and an IV has been used that is moderately or strongly related to exposure, it is likely that the assumptions of IV are violated, resulting in a biased effect estimate. We conclude that instrumental variables can be useful in case of moderate confounding but are less useful when strong confounding exists, because strong instruments cannot be found and assumptions will be easily violated. (pidemiology 006;17: 60 67) In medical research, randomized, controlled trials (RCTs) remain the gold standard in assessing the effect of one variable of interest, often a specified treatment. Nevertheless, observational studies are often used in estimating such an effect. 1 In epidemiologic as well as sociologic and economic Submitted 8 February 005; accepted 16 November 005. From the *Department of Pharmacoepidemiology and Pharmacotherapy, Utrecht Institute of Pharmaceutical Sciences (UIPS), Utrecht University, Utrecht, The Netherlands; and the Centre for Biostatistics, Utrecht University, Utrecht, The Netherlands. Supported by the Utrecht Institute of Pharmaceutical Sciences (UIPS), Utrecht University, Utrecht, The Netherlands. Correspondence: Olaf H. Klungel, Department of Pharmacoepidemiology and Pharmacotherapy, Utrecht Institute of Pharmaceutical Sciences (UIPS), Utrecht University, Sorbonnelaan 16, 3584 CA Utrecht, The Netherlands. -mail: o.h.klungel@pharm.uu.nl. Copyright 006 by Lippincott Williams & Wilkins ISSN: /06/ DOI: /01.ede cb 60 research, observational studies are the standard for exploring causal relationships between an exposure and an outcome variable. The main problem of estimating the effect in such studies is the potential bias resulting from confounding between the variable of interest and alternative explanations for the outcome (confounders). Traditionally, standard methods such as stratification, matching, and multiple regression techniques have been used to deal with confounding. In the epidemiologic literature, some other methods have been proposed,3 of which the method of propensity scores is best known. 4 In most of these methods, adjustment can be made only for observed confounders. A method that has the potential to adjust for all confounders, whether observed or not, is the method of instrumental variables (IV). This method is well known in economics and econometrics as the estimation of simultaneous regression equations 5 and is also referred to as structural equations and two-stage least squares. This method has a long tradition in economic literature, but has entered more recently into the medical research literature with increased focus on the validity of the instruments. Introductory texts on instrumental variables can be found in Greenland 6 and ohoori and Savitz. 7 One of the earliest applications of IV in the medical field is probably the research of Permutt and Hebel, 8 who estimated the effect of smoking of pregnant women on their child s birth weight, using an encouragement to stop smoking as the instrumental variable. More recent examples can be found in Beck et al, 9 Brooks et al, 10 arle et al, 11 Hadley et al, 1 Leigh and Schembri, 13 McClellan, 14 and McIntosh. 15 However, it has been argued that the application of this method is limited because of its strong assumptions, making it difficult in practice to find a suitable instrumental variable. 16 The objectives of this article are first to introduce the application of the method of IV in epidemiology in a nontechnical way and second, to show the limitations of this method, from which it follows that IV is less useful for solving large confounding problems such as confounding by indication. A SIMPL LINAR INSTRUMNTAL VARIABLS MODL In an RCT, the main purpose is to estimate the effect of one explanatory factor (the treatment) on an outcome variable. Because treatments have been randomly assigned to individuals, the treatment variable is in general independent pidemiology Volume 17, Number 3, May 006
2 pidemiology Volume 17, Number 3, May 006 of other explanatory factors. In case of a continuous outcome and a linear model, this randomization procedure allows one to estimate the treatment effect by means of ordinary least squares with a well-known unbiased estimator (see, for instance, Pestman 17 ). In observational studies, on the other hand, one has no control over this explanatory factor (further denoted as exposure) so that ordinary least squares as an estimation method will generally be biased because of the existence of unmeasured confounders. For example, one cannot directly estimate the effect of cigarette smoking on health without considering confounding factors such as age and socioeconomic position. One way to adjust for all possible confounding factors, whether observed or not, is to make use of an instrumental variable. The idea is that the causal effect of exposure on outcome can be captured by using the relationship between the exposure and another variable, the instrumental variable. How this variable can be selected and which conditions have to be fulfilled is discussed subsequently. First, we illustrate the model and its estimator. The Model and Its stimator A simple linear model for IV estimation consists of equations: Y (1) F () where Y is the outcome variable, is the exposure, is the instrumental variable, and and F are errors. In this set of structural equations, the variable is endogenous, which means that it is explained by other variables in the model, in this case the instrumental variable. is supposed to be linearly related to and exogenous, ie, explained by variables outside the model. For simplicity, we restrict ourselves to one instrumental variable, equations, and no other explaining variables. Under conditions further outlined in the next section, it can be proved that equation (3) presents an asymptotically unbiased estimate of the effect of on Y 18 : ˆ iv 1 n n 1 i 1 1 n n 1 i 1 (z i z )(y i y ) (z i z ) x i x ) ˆ,Y (3) ˆ, where ˆ,Y is the sample covariance of and Y and ˆ, is the sample covariance of and. It is more convenient to express the IV estimator in terms of ordinary least squares estimators: ˆ iv ˆ,Y ˆ,Y ˆ ˆ, ˆ, ˆ ˆ ols( Y) ˆ ols( ) (4) The numerator equals the effect of the instrumental variable on the outcome, whereas in the denominator, the effect of the IV on the exposure is given. In case of a dichotomous IV, the numerator equals simply the difference in mean outcome between 0 and 1 and the denominator equals the difference in mean exposure. When the outcome and exposure variable are also dichotomous and linearity is still assumed, this model is known as a linear probability model. In that case, the IV estimator presented here can be simply expressed as probabilities 18 : P(Y 1 1) P(Y 1 0) ˆ iv (5) P( 1 1) P( 1 0) where P(Y 1 1) P(Y 1 0) equals the risk difference of an event between 1 and 0. How to Obtain a Valid Instrumental Variable One can imagine that a method that claims to adjust for all possible confounders without randomization of treatments puts high requirements on the IV to be used for estimation. When this method is applied, 3 important assumptions have been made. The first assumption is the existence of at least some correlation between the IV and the exposure, because otherwise, equation () would be useless and the denominator of equation (4) would be equal to zero. In addition to this formal condition, it is important that this correlation should not be too small (see Implications of Weak Instruments ). The second assumption is that the relationship between the instrumental variable and the exposure is not confounded by other variables so that equation () is estimated without bias. This is the same as saying that the correlation between the IV and the error F must be equal to zero. One way to achieve this is to use as IV a variable that is controlled by the researcher. An example can be found in Permutt and Hebel, 8 in which a randomized encouragement to stop smoking was used as the IV to estimate the effect of smoking by pregnant women on child s birth weight. The researchers used encouragement regimes, an encouragement to stop smoking versus no encouragement, randomly assigned to pregnant smoking women. Alternatively, in some situations, a natural randomization process can be used as the IV. An an example, also known as Mendelian randomization, can be found in genetics in which alleles are considered to be allocated at random in offspring with the same parents. 19,0 In a study on the causality between low serum cholesterol and cancer, a genetic determinant of serum cholesterol was used as the instrumental variable. 1, When neither an active randomization nor a natural randomization is feasible to obtain an IV, the only possibility is to select an IV on theoretical grounds, assuming and reasoning that the relationship between the IV and the exposure can be estimated without bias. Such an example can be found in Leigh and Schembri 13 in which the observed cigarette price per region was used as the IV in a study on the relationship between smoking and health. The authors argued that there was no bias in estimating the relationship between cigarette price and smoking because the price elasticities in their study (the percentage change in 006 Lippincott Williams & Wilkins 61
3 Martens et al pidemiology Volume 17, Number 3, May 006 number of cigarettes smoked related to the percentage change in cigarette price) matched the price elasticities mentioned in the literature. The third assumption for an IV is most crucial and states that there should be no correlation between the IV and the error (further referred to as the main assumption). This means that the instrumental variable should influence the outcome neither directly nor indirectly by its relationship with other variables. Whether this assumption is valid can be argued only theoretically, and cannot be tested empirically. These 3 assumptions can be summarized as follows: 1., 0, no zero-correlation between IV and exposure;.,f 0, no correlation between IV and other factors explaining (error F); and 3., 0, no correlation between IV and other factors explaining Y (error ), main assumption. It should be noted that confounders of the -Y relation are not explicitly mentioned in these assumptions and that these confounders are part of both errors and F. In the special case that,f 1, the assumption could be formulated by referring to confounders only. 6 Numeric xample of Instrumental Variable Application As an example of IV estimation, we use the research of Permutt and Hebel. 8 Here the effect of smoking () by pregnant women on child s birth weight (Y) was studied. The instrumental variable () was the randomization procedure used to assign women to an encouragement program to stop smoking, which fulfills the second assumption. To apply IV estimation, first the intention-to-treat estimator ols(3y) needs to be calculated. In case of a dichotomous IV, this simply equals the difference in mean birth weight between women who were encouraged to stop smoking and women who were not ( ols(3y) 98 g). Next, we calculate the difference between encouragement groups in the fraction of women who stopped smoking ( ols(3) ). The ratio equals the IV estimator 98/( ) 430 g, indicating that stopping smoking raises average birth weight by 430 g. Figure 1 illustrates this calculation, in which actually stopped smoking is denoted as 1 and continued to smoke as 0. The encouragement smoking relationship and the encouragement birth weight relationship are represented by the solid lines in the lower and upper panel, respectively. Under the assumptions of IV estimation, the effect of smoking on birth weight is known only when smoking is changed from 0.43 to 0.0, in which in fact interest is in a change from 0to 1. xtending this difference to a difference from 0 to 1, indicated by the dotted line in the lower panel, and using the relationship between and Y in the upper panel, the intention-to-treat estimator of 98 g is extended to become the IV estimator of 430 g. Reminding that our second assumption has been fulfilled by randomization, the possible bias of the IV estimator mainly depends on the assumption that there should be no effect from encouragement on child s birth weight other than by means of changing smoking behavior. Such an effect cannot be ruled out completely, for 6 FIGUR 1. The instrumental variable estimator in the study of Permutt and Hebel. 8 instance, because women who were encouraged to stop smoking could become also more motivated to change other health-related behavior as well (for instance, nutrition). Birth weight will then be influenced by encouragement independently of smoking, which will lead to an overestimation of the effect of stopping smoking. IMPLICATIONS OF WAK INSTRUMNTS In the previous sections, the method and application of instrumental variables in a linear model were introduced in a nontechnical way. Here we focus on the implications when the correlation between the instrumental variable and the exposure is small or when the instrument is weak. We refer to this correlation as,. Large Standard rror A weak instrument means that the denominator in equation (4) is small. The smaller this covariance, the more sensitive the IV estimate will be to small changes. This sensitivity is mentioned by various authors 16,3 and can be deduced from the formula for the standard error: ˆ iv (6), where is the standard deviation of, is the standard deviation of, and, is the covariance of and. This covariance in the denominator behaves as a multiplier, which means that a small covariance (and hence a small correlation) will lead to a large standard error. In Figure 1, this sensitivity is reflected by the fact that the slope estimate in the lower 006 Lippincott Williams & Wilkins
4 pidemiology Volume 17, Number 3, May 006 panel becomes less reliable when the difference in between 0 and 1 becomes smaller. Bias When Sample Size Is Small An important characteristic of an estimator is that it should equal on average the true value (unbiasedness). Assuming that the assumptions of IV are not violated, the IV estimator is only asymptotically unbiased, meaning that on average bias will exist when the estimator iv is used in smaller samples. This bias appears because the relationship between the instrumental variable and the exposure is in general unknown and has to be estimated by equation (). As is usual in regression, overfitting generates a bias that depends on both the sample size and the correlation between the IV and the exposure. With moderate sample size and a weak instrument, this bias can become substantial. 4 It can be shown that this bias will be in the direction of the ordinary least squares estimator ols calculated in the simple linear regression of outcome on exposure. 3,5 Information on the magnitude of the small sample bias is contained in the F-statistic of the regression in equation (), which can be expressed as F ˆ, (n ) (7) 1 ˆ, An F-value not far from 1 indicates a large small sample bias, whereas a value of 10 seems to be sufficient for the bias to be negligible. 16 For example, in a sample of 50 independent observations, the correlation between and should be at least 0.0 to reach an F-value of 10. Another solution to deal with possible small sample bias is to use other IV estimators. 16,6 Bias When the Main Assumption Is Only Slightly Violated very violation of the main assumption of IV will naturally result in a biased estimator. More interesting is that only a small violation of this assumption will result in a large bias in case of a weak instrument because of its multiplicative effect in the estimator. Bound et al 3 expressed this bias in infinitely large samples (inconsistency) as a relative measure compared with the bias in the ordinary least squares estimator lim ˆ iv,/, lim ˆ ols (8), where lim is the limit as sample size increases. From this formula, it can be seen that even a small correlation between the instrumental variable and the error (, in the numerator) will produce a large inconsistency in the IV estimate relative to the ordinary least squares estimate when the instrument is weak, ie, when, is small. Thus, when has some small direct effect on Y, or an indirect effect other than through, the IV estimate will be increasingly biased when the instrument becomes weaker, even in very large samples. It can be concluded that a small correlation between the IV and the exposure can be a threat for the validity of the IV method, mainly in combination with a small sample or a possible violation of the main assumption. Although known from the literature, this aspect is often overlooked. A LIMIT ON TH STRNGTH OF INSTRUMNTS From the last section, it follows that the correlation between a possible instrumental variable and exposure (the strength of the IV, ) has to be as strong as possible, which also intuitively makes sense. However, in practice, it is often difficult to obtain an IV that is strongly related to exposure. One reason can be found in the existence of an upper bound on this correlation, which depends on the amount of confounding (indicated by, ), the correlation between the errors in the model (,F ), and the degree of violation of the main assumption (, ). We further explore the relationship between these correlations and distinguish between a situation in which the main assumption is fulfilled and one in which it is not. When the Main Assumption Has Been Fulfilled In case the main assumption of IV has been fulfilled, which means that the IV changes the outcome only through its relationship with the exposure, it can be shown that, 1, (9),F of which the proof is given in Appendix A. quation (9) indicates that there is a maximum on the strength of the instrumental variable and that this maximum decreases when the amount of confounding increases. In case of considerable confounding, the maximum correlation between IV and exposure will be quite low. This relationship between the correlations is illustrated in Figure. The relation between the strength of the IV, and the amount of confounding, is illustrated by curves representing various levels of the correlation between the errors,f. It can be seen that the maximum correlation between the potential instrumental variable and exposure becomes smaller when the amount of confounding becomes larger. When, for example, there is considerable confounding by indication (, 0.8), the maximum strength of the IV is 0.6. Probably this maximum will be even lower because the correlation between the errors will generally be less than 1.0. When, for instance,,f 0.85, this maximum drops to only Of the 3 correlations presented in equation (9) and Figure, the correlation between the errors is most difficult to understand. For the main message, however, its existence is not essential, as is illustrated in Figure 3 using vectors. In Figure 3A, the angle between and is close to 90, meaning that their correlation is small (small confounding). Because has to be uncorrelated with according to the third IV assumption (perpendicular), the angle between and will be automatically small, indicating a strong IV. In contrast, Figure 3B shows that a large confounding problem (small angle between and ) implies a weak instrument (large angle and small correlation between and ). The tradeoff between these correlations is an important characteristic of IV estimation. (Note that we simplified the figure by 006 Lippincott Williams & Wilkins 63
5 Martens et al pidemiology Volume 17, Number 3, May 006 FIGUR. Relationship between strength of an instrumental variable (, ) and amount of confounding (, ) for different error correlation levels (,F ) when main assumption has been fulfilled (, 0). a FIGUR 3. Relationship among,, and expressed in vectors. 64 b choosing in the same plane as and Y to remove,f from the figure because it equals its maximum of 1.0. See Appendix B for the situation in which is not in this plane.) As has been said, the correlation between the errors,f also plays a role. To better understand its meaning, we give examples. In Permutt and Hebel, 8 it is likely that this correlation will be small. Other reasons for birth weight variation besides smoking include socioeconomic conditions, inadequate nutrition, abuse, genetic factors, ethnic factors, physical work conditions, and chronic diseases. Because these explanatory factors for birth weight will be only partly overlapping with the reasons for noncompliance, ie, to continue smoking while encouraged to stop,,f is expected to be small. When, on the other hand, this correlation approaches 1, it means that the set of variables accounting for the unexplained variation in the outcome Y (error ) is strongly correlated with the unexplained instrumental variance (error F). An example of such a large correlation is a case of strong confounding by indication, in which unobserved health problems are the main reason for getting an illness and also for receiving preventive treatment. That causes variables and F to be strongly correlated and the maximum strength of the IV to be relatively small (see the right side of Fig. ). When the Main Assumption Has Not Been Fulfilled When the main assumption has not been (completely) fulfilled, the correlation between and is not equal to 0. Because the correlation between the errors plays a minor role, this correlation has been set to its maximum value of 1. In that case, the next inequality holds:,,, 1, 1, (10) Like equation (9), this expression states that in case of considerable confounding, the strength of the instrumental variable is bound to a relatively small value. It further states that a tradeoff exists between, and, : given a certain degree of confounding, the strength of the IV can be enlarged by relaxing the main assumption. In practice, this means that when IV is applied to a situation in which a considerable amount of confounding is to be expected and a very strong instrument has been found, it is very likely that the main assumption has been violated. The ffect on Bias The limit of the correlation between exposure and instrumental variable has an indirect effect on the bias, because the correlation to be found in practice will be low. This has several disadvantages that can be illustrated using some previous numeric examples. Suppose we deal with 006 Lippincott Williams & Wilkins
6 pidemiology Volume 17, Number 3, May 006 strong confounding by indication, say, As has been argued before, this will naturally imply a strong but imperfect correlation between the errors, say,f In that case, the limit of the correlation between exposure and IV will be, Restricting ourselves to instrumental variables that fulfill the main assumption (, 0), it will be practically impossible to find an IV that possesses the characteristic of being maximally correlated with exposure, which implies that this correlation will be lower than 0.34, for instance 0.0. With such a small correlation, the effect on the bias will be substantial when sample size falls below 50 observations. Because we cannot be sure that the main assumption has been fulfilled, care must be taken even with larger samples sizes. DISCUSSION We have focused on the method of instrumental variables for its ability to adjust for confounding in nonrandomized studies. We have explained the method and its application in a linear model and focused on the correlation between the IV and the exposure. When this correlation is very small, this method will lead to an increased standard error of the estimate, a considerable bias when sample size is small, and a bias even in large samples when the main assumption is only slightly violated. Furthermore, we demonstrated the existence of an upper bound on the correlation between the IV and the exposure. This upper bound is not a practical limitation when confounding is small or moderate because the maximum strength of the IV is still very high. When, on the other hand, considerable confounding by indication exists, the maximum correlation between any potential IV and the exposure will be quite low, resulting possibly in a fairly weak instrument to fulfill the main assumption. Because of a tradeoff between violation of this main assumption and the strength of the IV, the presence of considerable confounding and a strong instrument will probably indicate a violation of the main assumption and thus a biased estimate. This article serves as an introduction on the method of instrumental variables demonstrating its merits and limitations. Complexities such as more equations, more instruments, the inclusion of covariates, and nonlinearity of the model have been left out. More equations could be added with more than endogenous variables, although it is unlikely to be useful in epidemiology when estimating an exposure (treatment) effect. In equation (), multiple instruments could be used; this extension does not change the basic ideas behind this method. 7 An advantage of more than one instrumental variable is that a test on the exogeneity of the instruments is possible. 16 Another extension is the inclusion of measured covariates in both equations. 7 We limited the model to linear regression, assuming that the outcome and the exposure are both continuous variables, while in medical research, dichotomous outcomes or exposures are more common. The main reason for this choice is simplicity: the application and implications can be more easily presented in a linear framework. A dichotomous outcome or dichotomous exposure can easily fit into this model when linearity is assumed using a linear probability model. Although less known, the results from this model are practically indistinguishable from logistic and probit regression analyses as long as the estimated probabilities range between 0. and ,9 When risk ratios or log odds are to be analyzed, like in logistic regression analysis, the presented IV estimator cannot be used and more complex IV estimators are required. We refer to the literature for IV estimation in such cases or in nonlinear models in general. 6,30,31 The limitations when instruments are weak, and the impossibility of finding strong instruments in the presence of strong confounding, apply in a similar way. When assessing the validity of study results, investigators should report both the correlation between IV and exposure (or difference in means) and the F-value resulting from equation () and given in equation (7). When either of these is small, instrumental variables will not produce unbiased and reasonably precise estimates of exposure effect. Furthermore, it should be made clear whether the IV is randomized by the researcher, randomized by nature, or is simply an observed variable. In the latter case, evidence should be given that the various categories of the instrumental variable have similar distributions on important characteristics. Additionally, the assumption that the IV determines outcome only by means of exposure is crucial. Because this cannot be checked, it should be argued theoretically that a direct or indirect relationship between the IV and the outcome is negligible. Finally, in a study in which considerable confounding can be expected (eg, strong confounding by indication), one should be aware that the existence of a very strong instrument within the IV assumptions is impossible. Whether the instrument is sufficiently correlated with exposure depends on the number of observations and the plausibility of the main assumption. We conclude that the method of IV can be useful in case of moderate confounding but is less useful when strong confounding (by indication) exists, because strong instruments cannot be found and assumptions will be easily violated. RFRNCS 1. Concato J, Shah N, Horwitz RI. Randomized, controlled trials, observational studies, and the hierarchy of research designs. N ngl J Med. 000;34: McMahon AD. Approaches to combat with confounding by indication in observational studies of intended drug effects. Pharmacoepidemiol Drug Saf. 003;1: Klungel OH, Martens P, Psaty BM, et al. Methods to assess intended effects of drug treatment in observational studies are reviewed. J Clin pidemiol. 004;57: Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika. 1983;70: Theil H. Principles of conometrics. New York: Wiley; Greenland S. An introduction to instrumental variables for epidemiologists. Int J pidemiol. 000;9: ohoori N, Savitz DA. conometric approaches to epidemiologic data: relating endogeneity and unobserved heterogeneity to confounding. Ann pidemiol. 1997;7: rratum in Ann pidemiol. 1997;7: Permutt TH, Hebel JR. Simultaneous-equation estimation in a clinical trial of the effect of smoking on birth weight. Biometrics. 1989;45: Beck CA, Penrod J, Gyorkos TW, et al. Does aggressive care following acute myocardial infarction reduce mortality? Analysis with instrumental variables to compare effectiveness in Canadian and United States patient populations. Health Serv Res. 003;38: Lippincott Williams & Wilkins 65
7 Martens et al pidemiology Volume 17, Number 3, May Brooks JM, Chrischilles A, Scott SD, et al. Was breast conserving surgery underutilized for early stage breast cancer? Instrumental variables evidence for stage II patients from Iowa. Health Serv Res. 003; 38: rratum in Health Serv Res. 004;39: arle CC, Tsai JS, Gelber RD, et al. ffectiveness of chemotherapy for advanced lung cancer in the elderly: instrumental variable and propensity analysis. J Clin Oncol. 001;19: Hadley J, Polsky D, Mandelblatt JS, et al. An exploratory instrumental variable analysis of the outcomes of localized breast cancer treatments in a medicare population. Health con. 003;1: Leigh JP, Schembri M. Instrumental variables technique: cigarette price provided better estimate of effects of smoking on SF-1. J Clin pidemiol. 004;57: McClellan M, McNeil BJ, Newhouse JP. Does more intensive treatment of acute myocardial infarction in the elderly reduce mortality? Analysis using instrumental variables. JAMA. 1994;7: McIntosh MW. Instrumental variables when evaluating screening trials: estimating the benefit of detecting cancer by screening. Stat Med. 1999;18: Staiger D, Stock JH. Instrumental variables regression with weak instruments. conometrica. 1997;65: Pestman WR. Mathematical Statistics. Walter de Gruyter, Angrist JD, Imbens GW, Rubin DB. Identification of causal effects using instrumental variables. JASA. 1996;91: Thomas DC, Conti DV. Commentary: the concept of mendelian randomization. Int J pidemiol. 004;33: Minelli C, Thompson JR, Tobin MD, et al. An integrated approach to the meta-analysis of genetic association studies using mendelian randomization. Am J pidemiol. 004;160: Katan MB. Apolipoprotein isoforms, serum cholesterol, and cancer. Lancet. 1986;1: Smith GD, brahim S. Mendelian randomization: prospects, potentials, and limitations. Int J pidemiol. 004;33: Bound J, Jaeger DA, Baker RM. Problems with instrumental variables estimation when the correlation between the instruments and the endogenous explanatory variable is weak. JASA. 1995;90: Sawa T. The exact sampling distribution of ordinary least squares and two-stage least squares estimators. J Am Stat Assoc. 1969;64: Nelson CR, Startz R. Some further results on the exact small sample properties of the instrumental variable estimator. conometrica. 1990;58: Angrist JD, Krueger AB. Split sample instrumental variables. J Bus con Stat. 1995;13: Angrist JD, Imbens GW. Two-stage least squares estimation of average causal effects in models with variable treatment intensity. JASA. 1995;90: Cox DR, Snell J. Analysis of Binary Data. Chapman and Hall, Cox DR, Wermuth N. A comment on the coefficient of determination for binary responses. American Statistician. 199;46: Bowden RJ, Turkington DA. A comparative study of instrumental variables estimators for nonlinear simultaneous models. J Am Stat Assoc. 1981;76: Amemiya T. The nonlinear two-stage least-squares estimator. Journal of conometrics. 1974;: It follows from this that,, F, 0 0,F,F. Using this expression for,, one derives that,,,f F F,F F F,F,F(1, ) Squaring, rearranging terms, and taking square roots will give which proves the theorem., 1,,F APPNDI B The condition,f 1 is equivalent to the condition that is in the same plane as and as can be seen in Figure 4. For simplicity, we assume that the expectation values of the variables, Y, and are all equal to zero. a FIGUR 4. Relationship among,,, and F expressed in vectors. ' b F APPNDI A Theorem 1 The correlation between and,, is bound to obey the equality, 1, Proof: According to the model, one has { Y F with, 0 and,f 0 66 (11),F V O ' FIGUR 5. Three-dimensional picture of,,, and noise O expressed in vectors. 006 Lippincott Williams & Wilkins
8 pidemiology Volume 17, Number 3, May 006 According to the IV condition that, 0 (these are perpendicular in panel a) and the condition that,f 0, it follows from panel b that and F necessarily point in the same or opposite direction, implying,f 1. In this situation, there is (up to scalar multiples) only one instrumental variable possible in the plane spanned by and. Ashas been argued in the text, it is not likely that this correlation equals 1. This is visualized in Figure 5 in which is not in the plane spanned by and, meaning that F, which is in the plane spanned by and and perpendicular to, can impossibly point in the same direction as. Consequently, one then has,f 1. Here is the projection of on the plane spanned by and. The vector can now be decomposed as O where is in the plane spanned by and and where O is perpendicular to this plane. The vector O can be referred to as noise because it is uncorrelated to both and Y. Note that the variable is an instrumental variable itself. 006 Lippincott Williams & Wilkins 67
Brief introduction to instrumental variables. IV Workshop, Bristol, Miguel A. Hernán Department of Epidemiology Harvard School of Public Health
Brief introduction to instrumental variables IV Workshop, Bristol, 2008 Miguel A. Hernán Department of Epidemiology Harvard School of Public Health Goal: To consistently estimate the average causal effect
More informationSample size and power calculations in Mendelian randomization with a single instrumental variable and a binary outcome
Sample size and power calculations in Mendelian randomization with a single instrumental variable and a binary outcome Stephen Burgess July 10, 2013 Abstract Background: Sample size calculations are an
More informationPropensity Score Analysis Shenyang Guo, Ph.D.
Propensity Score Analysis Shenyang Guo, Ph.D. Upcoming Seminar: April 7-8, 2017, Philadelphia, Pennsylvania Propensity Score Analysis 1. Overview 1.1 Observational studies and challenges 1.2 Why and when
More informationMethods for Addressing Selection Bias in Observational Studies
Methods for Addressing Selection Bias in Observational Studies Susan L. Ettner, Ph.D. Professor Division of General Internal Medicine and Health Services Research, UCLA What is Selection Bias? In the regression
More informationCitation for published version (APA): Ebbes, P. (2004). Latent instrumental variables: a new approach to solve for endogeneity s.n.
University of Groningen Latent instrumental variables Ebbes, P. IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document
More informationInstrumental Variables Estimation: An Introduction
Instrumental Variables Estimation: An Introduction Susan L. Ettner, Ph.D. Professor Division of General Internal Medicine and Health Services Research, UCLA The Problem The Problem Suppose you wish to
More informationWrite your identification number on each paper and cover sheet (the number stated in the upper right hand corner on your exam cover).
STOCKHOLM UNIVERSITY Department of Economics Course name: Empirical methods 2 Course code: EC2402 Examiner: Per Pettersson-Lidbom Number of credits: 7,5 credits Date of exam: Sunday 21 February 2010 Examination
More informationSUPPLEMENTARY INFORMATION
Supplementary Statistics and Results This file contains supplementary statistical information and a discussion of the interpretation of the belief effect on the basis of additional data. We also present
More informationEMPIRICAL STRATEGIES IN LABOUR ECONOMICS
EMPIRICAL STRATEGIES IN LABOUR ECONOMICS University of Minho J. Angrist NIPE Summer School June 2009 This course covers core econometric ideas and widely used empirical modeling strategies. The main theoretical
More informationCan you guarantee that the results from your observational study are unaffected by unmeasured confounding? H.Hosseini
Instrumental Variable (Instrument) applications in Epidemiology: An Introduction Hamed Hosseini Can you guarantee that the results from your observational study are unaffected by unmeasured confounding?
More informationInstrumental Variables I (cont.)
Review Instrumental Variables Observational Studies Cross Sectional Regressions Omitted Variables, Reverse causation Randomized Control Trials Difference in Difference Time invariant omitted variables
More informationCausal Validity Considerations for Including High Quality Non-Experimental Evidence in Systematic Reviews
Non-Experimental Evidence in Systematic Reviews OPRE REPORT #2018-63 DEKE, MATHEMATICA POLICY RESEARCH JUNE 2018 OVERVIEW Federally funded systematic reviews of research evidence play a central role in
More informationPropensity Score Methods for Estimating Causality in the Absence of Random Assignment: Applications for Child Care Policy Research
2012 CCPRC Meeting Methodology Presession Workshop October 23, 2012, 2:00-5:00 p.m. Propensity Score Methods for Estimating Causality in the Absence of Random Assignment: Applications for Child Care Policy
More informationLecture II: Difference in Difference. Causality is difficult to Show from cross
Review Lecture II: Regression Discontinuity and Difference in Difference From Lecture I Causality is difficult to Show from cross sectional observational studies What caused what? X caused Y, Y caused
More informationA Brief Introduction to Bayesian Statistics
A Brief Introduction to Statistics David Kaplan Department of Educational Psychology Methods for Social Policy Research and, Washington, DC 2017 1 / 37 The Reverend Thomas Bayes, 1701 1761 2 / 37 Pierre-Simon
More informationDylan Small Department of Statistics, Wharton School, University of Pennsylvania. Based on joint work with Paul Rosenbaum
Instrumental variables and their sensitivity to unobserved biases Dylan Small Department of Statistics, Wharton School, University of Pennsylvania Based on joint work with Paul Rosenbaum Overview Instrumental
More informationPros. University of Chicago and NORC at the University of Chicago, USA, and IZA, Germany
Dan A. Black University of Chicago and NORC at the University of Chicago, USA, and IZA, Germany Matching as a regression estimator Matching avoids making assumptions about the functional form of the regression
More informationSupplement 2. Use of Directed Acyclic Graphs (DAGs)
Supplement 2. Use of Directed Acyclic Graphs (DAGs) Abstract This supplement describes how counterfactual theory is used to define causal effects and the conditions in which observed data can be used to
More informationQuantitative Methods. Lonnie Berger. Research Training Policy Practice
Quantitative Methods Lonnie Berger Research Training Policy Practice Defining Quantitative and Qualitative Research Quantitative methods: systematic empirical investigation of observable phenomena via
More informationInstrumental variable analysis in randomized trials with noncompliance. for observational pharmacoepidemiologic
Page 1 of 7 Statistical/Methodological Debate Instrumental variable analysis in randomized trials with noncompliance and observational pharmacoepidemiologic studies RHH Groenwold 1,2*, MJ Uddin 1, KCB
More informationWhat is Multilevel Modelling Vs Fixed Effects. Will Cook Social Statistics
What is Multilevel Modelling Vs Fixed Effects Will Cook Social Statistics Intro Multilevel models are commonly employed in the social sciences with data that is hierarchically structured Estimated effects
More informationMarno Verbeek Erasmus University, the Netherlands. Cons. Pros
Marno Verbeek Erasmus University, the Netherlands Using linear regression to establish empirical relationships Linear regression is a powerful tool for estimating the relationship between one variable
More informationApplied Quantitative Methods II
Applied Quantitative Methods II Lecture 7: Endogeneity and IVs Klára Kaĺıšková Klára Kaĺıšková AQM II - Lecture 7 VŠE, SS 2016/17 1 / 36 Outline 1 OLS and the treatment effect 2 OLS and endogeneity 3 Dealing
More informationCausal Mediation Analysis with the CAUSALMED Procedure
Paper SAS1991-2018 Causal Mediation Analysis with the CAUSALMED Procedure Yiu-Fai Yung, Michael Lamm, and Wei Zhang, SAS Institute Inc. Abstract Important policy and health care decisions often depend
More informationSensitivity Analysis in Observational Research: Introducing the E-value
Sensitivity Analysis in Observational Research: Introducing the E-value Tyler J. VanderWeele Harvard T.H. Chan School of Public Health Departments of Epidemiology and Biostatistics 1 Plan of Presentation
More informationIdentification of population average treatment effects using nonlinear instrumental variables estimators : another cautionary note
University of Iowa Iowa Research Online Theses and Dissertations Fall 2014 Identification of population average treatment effects using nonlinear instrumental variables estimators : another cautionary
More informationImproved control for confounding using propensity scores and instrumental variables?
Improved control for confounding using propensity scores and instrumental variables? Dr. Olaf H.Klungel Dept. of Pharmacoepidemiology & Clinical Pharmacology, Utrecht Institute of Pharmaceutical Sciences
More informationCase A, Wednesday. April 18, 2012
Case A, Wednesday. April 18, 2012 1 Introduction Adverse birth outcomes have large costs with respect to direct medical costs aswell as long-term developmental consequences. Maternal health behaviors at
More informationSome interpretational issues connected with observational studies
Some interpretational issues connected with observational studies D.R. Cox Nuffield College, Oxford, UK and Nanny Wermuth Chalmers/Gothenburg University, Gothenburg, Sweden ABSTRACT After some general
More informationMEA DISCUSSION PAPERS
Inference Problems under a Special Form of Heteroskedasticity Helmut Farbmacher, Heinrich Kögel 03-2015 MEA DISCUSSION PAPERS mea Amalienstr. 33_D-80799 Munich_Phone+49 89 38602-355_Fax +49 89 38602-390_www.mea.mpisoc.mpg.de
More informationCochrane Pregnancy and Childbirth Group Methodological Guidelines
Cochrane Pregnancy and Childbirth Group Methodological Guidelines [Prepared by Simon Gates: July 2009, updated July 2012] These guidelines are intended to aid quality and consistency across the reviews
More informationPropensity Score Matching with Limited Overlap. Abstract
Propensity Score Matching with Limited Overlap Onur Baser Thomson-Medstat Abstract In this article, we have demostrated the application of two newly proposed estimators which accounts for lack of overlap
More informationThe Effects of Maternal Alcohol Use and Smoking on Children s Mental Health: Evidence from the National Longitudinal Survey of Children and Youth
1 The Effects of Maternal Alcohol Use and Smoking on Children s Mental Health: Evidence from the National Longitudinal Survey of Children and Youth Madeleine Benjamin, MA Policy Research, Economics and
More informationComplier Average Causal Effect (CACE)
Complier Average Causal Effect (CACE) Booil Jo Stanford University Methodological Advancement Meeting Innovative Directions in Estimating Impact Office of Planning, Research & Evaluation Administration
More informationIntroduction to Applied Research in Economics Kamiljon T. Akramov, Ph.D. IFPRI, Washington, DC, USA
Introduction to Applied Research in Economics Kamiljon T. Akramov, Ph.D. IFPRI, Washington, DC, USA Training Course on Applied Econometric Analysis June 1, 2015, WIUT, Tashkent, Uzbekistan Why do we need
More informationBias in regression coefficient estimates when assumptions for handling missing data are violated: a simulation study
STATISTICAL METHODS Epidemiology Biostatistics and Public Health - 2016, Volume 13, Number 1 Bias in regression coefficient estimates when assumptions for handling missing data are violated: a simulation
More informationChapter 11 Nonexperimental Quantitative Research Steps in Nonexperimental Research
Chapter 11 Nonexperimental Quantitative Research (Reminder: Don t forget to utilize the concept maps and study questions as you study this and the other chapters.) Nonexperimental research is needed because
More information11/18/2013. Correlational Research. Correlational Designs. Why Use a Correlational Design? CORRELATIONAL RESEARCH STUDIES
Correlational Research Correlational Designs Correlational research is used to describe the relationship between two or more naturally occurring variables. Is age related to political conservativism? Are
More informationEstimating Heterogeneous Choice Models with Stata
Estimating Heterogeneous Choice Models with Stata Richard Williams Notre Dame Sociology rwilliam@nd.edu West Coast Stata Users Group Meetings October 25, 2007 Overview When a binary or ordinal regression
More informationDetection of Unknown Confounders. by Bayesian Confirmatory Factor Analysis
Advanced Studies in Medical Sciences, Vol. 1, 2013, no. 3, 143-156 HIKARI Ltd, www.m-hikari.com Detection of Unknown Confounders by Bayesian Confirmatory Factor Analysis Emil Kupek Department of Public
More informationWRITTEN PRELIMINARY Ph.D. EXAMINATION. Department of Applied Economics. January 17, Consumer Behavior and Household Economics.
WRITTEN PRELIMINARY Ph.D. EXAMINATION Department of Applied Economics January 17, 2012 Consumer Behavior and Household Economics Instructions Identify yourself by your code letter, not your name, on each
More informationRussian Journal of Agricultural and Socio-Economic Sciences, 3(15)
ON THE COMPARISON OF BAYESIAN INFORMATION CRITERION AND DRAPER S INFORMATION CRITERION IN SELECTION OF AN ASYMMETRIC PRICE RELATIONSHIP: BOOTSTRAP SIMULATION RESULTS Henry de-graft Acquah, Senior Lecturer
More informationLogistic regression: Why we often can do what we think we can do 1.
Logistic regression: Why we often can do what we think we can do 1. Augst 8 th 2015 Maarten L. Buis, University of Konstanz, Department of History and Sociology maarten.buis@uni.konstanz.de All propositions
More informationTesting the Predictability of Consumption Growth: Evidence from China
Auburn University Department of Economics Working Paper Series Testing the Predictability of Consumption Growth: Evidence from China Liping Gao and Hyeongwoo Kim Georgia Southern University and Auburn
More informationLecture II: Difference in Difference and Regression Discontinuity
Review Lecture II: Difference in Difference and Regression Discontinuity it From Lecture I Causality is difficult to Show from cross sectional observational studies What caused what? X caused Y, Y caused
More informationPerformance of prior event rate ratio adjustment method in pharmacoepidemiology: a simulation study
pharmacoepidemiology and drug safety (2014) Published online in Wiley Online Library (wileyonlinelibrary.com).3724 ORIGINAL REPORT Performance of prior event rate ratio adjustment method in pharmacoepidemiology:
More informationGlossary From Running Randomized Evaluations: A Practical Guide, by Rachel Glennerster and Kudzai Takavarasha
Glossary From Running Randomized Evaluations: A Practical Guide, by Rachel Glennerster and Kudzai Takavarasha attrition: When data are missing because we are unable to measure the outcomes of some of the
More informationModule 14: Missing Data Concepts
Module 14: Missing Data Concepts Jonathan Bartlett & James Carpenter London School of Hygiene & Tropical Medicine Supported by ESRC grant RES 189-25-0103 and MRC grant G0900724 Pre-requisites Module 3
More informationReview: Logistic regression, Gaussian naïve Bayes, linear regression, and their connections
Review: Logistic regression, Gaussian naïve Bayes, linear regression, and their connections New: Bias-variance decomposition, biasvariance tradeoff, overfitting, regularization, and feature selection Yi
More informationIntroduction to Observational Studies. Jane Pinelis
Introduction to Observational Studies Jane Pinelis 22 March 2018 Outline Motivating example Observational studies vs. randomized experiments Observational studies: basics Some adjustment strategies Matching
More informationAdvanced IPD meta-analysis methods for observational studies
Advanced IPD meta-analysis methods for observational studies Simon Thompson University of Cambridge, UK Part 4 IBC Victoria, July 2016 1 Outline of talk Usual measures of association (e.g. hazard ratios)
More informationDoes Body Mass Index Adequately Capture the Relation of Body Composition and Body Size to Health Outcomes?
American Journal of Epidemiology Copyright 1998 by The Johns Hopkins University School of Hygiene and Public Health All rights reserved Vol. 147, No. 2 Printed in U.S.A A BRIEF ORIGINAL CONTRIBUTION Does
More informationEc331: Research in Applied Economics Spring term, Panel Data: brief outlines
Ec331: Research in Applied Economics Spring term, 2014 Panel Data: brief outlines Remaining structure Final Presentations (5%) Fridays, 9-10 in H3.45. 15 mins, 8 slides maximum Wk.6 Labour Supply - Wilfred
More informationJournal of Political Economy, Vol. 93, No. 2 (Apr., 1985)
Confirmations and Contradictions Journal of Political Economy, Vol. 93, No. 2 (Apr., 1985) Estimates of the Deterrent Effect of Capital Punishment: The Importance of the Researcher's Prior Beliefs Walter
More informationDichotomizing partial compliance and increased participant burden in factorial designs: the performance of four noncompliance methods
Merrill and McClure Trials (2015) 16:523 DOI 1186/s13063-015-1044-z TRIALS RESEARCH Open Access Dichotomizing partial compliance and increased participant burden in factorial designs: the performance of
More informationRecent developments for combining evidence within evidence streams: bias-adjusted meta-analysis
EFSA/EBTC Colloquium, 25 October 2017 Recent developments for combining evidence within evidence streams: bias-adjusted meta-analysis Julian Higgins University of Bristol 1 Introduction to concepts Standard
More informationINTRODUCTION TO ECONOMETRICS (EC212)
INTRODUCTION TO ECONOMETRICS (EC212) Course duration: 54 hours lecture and class time (Over three weeks) LSE Teaching Department: Department of Economics Lead Faculty (session two): Dr Taisuke Otsu and
More informationStatistical reports Regression, 2010
Statistical reports Regression, 2010 Niels Richard Hansen June 10, 2010 This document gives some guidelines on how to write a report on a statistical analysis. The document is organized into sections that
More informationCarrying out an Empirical Project
Carrying out an Empirical Project Empirical Analysis & Style Hint Special program: Pre-training 1 Carrying out an Empirical Project 1. Posing a Question 2. Literature Review 3. Data Collection 4. Econometric
More informationChoice of axis, tests for funnel plot asymmetry, and methods to adjust for publication bias
Technical appendix Choice of axis, tests for funnel plot asymmetry, and methods to adjust for publication bias Choice of axis in funnel plots Funnel plots were first used in educational research and psychology,
More informationPropensity score methods to adjust for confounding in assessing treatment effects: bias and precision
ISPUB.COM The Internet Journal of Epidemiology Volume 7 Number 2 Propensity score methods to adjust for confounding in assessing treatment effects: bias and precision Z Wang Abstract There is an increasing
More informationKey questions when starting an econometric project (Angrist & Pischke, 2009):
Econometric & other impact assessment approaches to policy analysis Part 1 1 The problem of causality in policy analysis Internal vs. external validity Key questions when starting an econometric project
More informationVersion No. 7 Date: July Please send comments or suggestions on this glossary to
Impact Evaluation Glossary Version No. 7 Date: July 2012 Please send comments or suggestions on this glossary to 3ie@3ieimpact.org. Recommended citation: 3ie (2012) 3ie impact evaluation glossary. International
More informationBIOSTATISTICAL METHODS AND RESEARCH DESIGNS. Xihong Lin Department of Biostatistics, University of Michigan, Ann Arbor, MI, USA
BIOSTATISTICAL METHODS AND RESEARCH DESIGNS Xihong Lin Department of Biostatistics, University of Michigan, Ann Arbor, MI, USA Keywords: Case-control study, Cohort study, Cross-Sectional Study, Generalized
More informationThe Impact of Relative Standards on the Propensity to Disclose. Alessandro Acquisti, Leslie K. John, George Loewenstein WEB APPENDIX
The Impact of Relative Standards on the Propensity to Disclose Alessandro Acquisti, Leslie K. John, George Loewenstein WEB APPENDIX 2 Web Appendix A: Panel data estimation approach As noted in the main
More informationStrategies for handling missing data in randomised trials
Strategies for handling missing data in randomised trials NIHR statistical meeting London, 13th February 2012 Ian White MRC Biostatistics Unit, Cambridge, UK Plan 1. Why do missing data matter? 2. Popular
More informationRegression Discontinuity Analysis
Regression Discontinuity Analysis A researcher wants to determine whether tutoring underachieving middle school students improves their math grades. Another wonders whether providing financial aid to low-income
More informationLecture Outline. Biost 590: Statistical Consulting. Stages of Scientific Studies. Scientific Method
Biost 590: Statistical Consulting Statistical Classification of Scientific Studies; Approach to Consulting Lecture Outline Statistical Classification of Scientific Studies Statistical Tasks Approach to
More informationSimulation study of instrumental variable approaches with an application to a study of the antidiabetic effect of bezafibrate
pharmacoepidemiology and drug safety 2012; 21(S2): 114 120 Published online in Wiley Online Library (wileyonlinelibrary.com).3252 ORIGINAL REPORT Simulation study of instrumental variable approaches with
More informationA COMPARISON OF IMPUTATION METHODS FOR MISSING DATA IN A MULTI-CENTER RANDOMIZED CLINICAL TRIAL: THE IMPACT STUDY
A COMPARISON OF IMPUTATION METHODS FOR MISSING DATA IN A MULTI-CENTER RANDOMIZED CLINICAL TRIAL: THE IMPACT STUDY Lingqi Tang 1, Thomas R. Belin 2, and Juwon Song 2 1 Center for Health Services Research,
More informationProblem Set 5 ECN 140 Econometrics Professor Oscar Jorda. DUE: June 6, Name
Problem Set 5 ECN 140 Econometrics Professor Oscar Jorda DUE: June 6, 2006 Name 1) Earnings functions, whereby the log of earnings is regressed on years of education, years of on-the-job training, and
More informationSelected Topics in Biostatistics Seminar Series. Missing Data. Sponsored by: Center For Clinical Investigation and Cleveland CTSC
Selected Topics in Biostatistics Seminar Series Missing Data Sponsored by: Center For Clinical Investigation and Cleveland CTSC Brian Schmotzer, MS Biostatistician, CCI Statistical Sciences Core brian.schmotzer@case.edu
More informationPLS 506 Mark T. Imperial, Ph.D. Lecture Notes: Reliability & Validity
PLS 506 Mark T. Imperial, Ph.D. Lecture Notes: Reliability & Validity Measurement & Variables - Initial step is to conceptualize and clarify the concepts embedded in a hypothesis or research question with
More informationAn Instrumental Variable Consistent Estimation Procedure to Overcome the Problem of Endogenous Variables in Multilevel Models
An Instrumental Variable Consistent Estimation Procedure to Overcome the Problem of Endogenous Variables in Multilevel Models Neil H Spencer University of Hertfordshire Antony Fielding University of Birmingham
More informationChapter 5: Field experimental designs in agriculture
Chapter 5: Field experimental designs in agriculture Jose Crossa Biometrics and Statistics Unit Crop Research Informatics Lab (CRIL) CIMMYT. Int. Apdo. Postal 6-641, 06600 Mexico, DF, Mexico Introduction
More informationQuasi-experimental analysis Notes for "Structural modelling".
Quasi-experimental analysis Notes for "Structural modelling". Martin Browning Department of Economics, University of Oxford Revised, February 3 2012 1 Quasi-experimental analysis. 1.1 Modelling using quasi-experiments.
More informationMendelian randomization with a binary exposure variable: interpretation and presentation of causal estimates
Mendelian randomization with a binary exposure variable: interpretation and presentation of causal estimates arxiv:1804.05545v1 [stat.me] 16 Apr 2018 Stephen Burgess 1,2 Jeremy A Labrecque 3 1 MRC Biostatistics
More informationCurrent Directions in Mediation Analysis David P. MacKinnon 1 and Amanda J. Fairchild 2
CURRENT DIRECTIONS IN PSYCHOLOGICAL SCIENCE Current Directions in Mediation Analysis David P. MacKinnon 1 and Amanda J. Fairchild 2 1 Arizona State University and 2 University of South Carolina ABSTRACT
More informationClass 1: Introduction, Causality, Self-selection Bias, Regression
Class 1: Introduction, Causality, Self-selection Bias, Regression Ricardo A Pasquini April 2011 Ricardo A Pasquini () April 2011 1 / 23 Introduction I Angrist s what should be the FAQs of a researcher:
More informationA Guide to Quasi-Experimental Designs
Western Kentucky University From the SelectedWorks of Matt Bogard Fall 2013 A Guide to Quasi-Experimental Designs Matt Bogard, Western Kentucky University Available at: https://works.bepress.com/matt_bogard/24/
More informationTHE USE OF MULTIVARIATE ANALYSIS IN DEVELOPMENT THEORY: A CRITIQUE OF THE APPROACH ADOPTED BY ADELMAN AND MORRIS A. C. RAYNER
THE USE OF MULTIVARIATE ANALYSIS IN DEVELOPMENT THEORY: A CRITIQUE OF THE APPROACH ADOPTED BY ADELMAN AND MORRIS A. C. RAYNER Introduction, 639. Factor analysis, 639. Discriminant analysis, 644. INTRODUCTION
More informationINTERNAL VALIDITY, BIAS AND CONFOUNDING
OCW Epidemiology and Biostatistics, 2010 J. Forrester, PhD Tufts University School of Medicine October 6, 2010 INTERNAL VALIDITY, BIAS AND CONFOUNDING Learning objectives for this session: 1) Understand
More informationSupplementary Appendix
Supplementary Appendix This appendix has been provided by the authors to give readers additional information about their work. Supplement to: Weintraub WS, Grau-Sepulveda MV, Weiss JM, et al. Comparative
More informationBayesian graphical models for combining multiple data sources, with applications in environmental epidemiology
Bayesian graphical models for combining multiple data sources, with applications in environmental epidemiology Sylvia Richardson 1 sylvia.richardson@imperial.co.uk Joint work with: Alexina Mason 1, Lawrence
More informationThe Limits of Inference Without Theory
The Limits of Inference Without Theory Kenneth I. Wolpin University of Pennsylvania Koopmans Memorial Lecture (2) Cowles Foundation Yale University November 3, 2010 Introduction Fuller utilization of the
More informationEXPERIMENTAL RESEARCH DESIGNS
ARTHUR PSYC 204 (EXPERIMENTAL PSYCHOLOGY) 14A LECTURE NOTES [02/28/14] EXPERIMENTAL RESEARCH DESIGNS PAGE 1 Topic #5 EXPERIMENTAL RESEARCH DESIGNS As a strict technical definition, an experiment is a study
More informationIdentifying Peer Influence Effects in Observational Social Network Data: An Evaluation of Propensity Score Methods
Identifying Peer Influence Effects in Observational Social Network Data: An Evaluation of Propensity Score Methods Dean Eckles Department of Communication Stanford University dean@deaneckles.com Abstract
More informationChallenges of Observational and Retrospective Studies
Challenges of Observational and Retrospective Studies Kyoungmi Kim, Ph.D. March 8, 2017 This seminar is jointly supported by the following NIH-funded centers: Background There are several methods in which
More informationCAN EFFECTIVENESS BE MEASURED OUTSIDE A CLINICAL TRIAL?
CAN EFFECTIVENESS BE MEASURED OUTSIDE A CLINICAL TRIAL? Mette Nørgaard, Professor, MD, PhD Department of Clinical Epidemiology Aarhus Universitety Hospital Aarhus, Denmark Danish Medical Birth Registry
More informationFlexible Matching in Case-Control Studies of Gene-Environment Interactions
American Journal of Epidemiology Copyright 2004 by the Johns Hopkins Bloomberg School of Public Health All rights reserved Vol. 59, No. Printed in U.S.A. DOI: 0.093/aje/kwg250 ORIGINAL CONTRIBUTIONS Flexible
More informationEstimating treatment effects with observational data: A new approach using hospital-level variation in treatment intensity
Preliminary and incomplete Do not quote Estimating treatment effects with observational data: A new approach using hospital-level variation in treatment intensity Mark McClellan Stanford University and
More informationApproaches to Improving Causal Inference from Mediation Analysis
Approaches to Improving Causal Inference from Mediation Analysis David P. MacKinnon, Arizona State University Pennsylvania State University February 27, 2013 Background Traditional Mediation Methods Modern
More informationEffects of propensity score overlap on the estimates of treatment effects. Yating Zheng & Laura Stapleton
Effects of propensity score overlap on the estimates of treatment effects Yating Zheng & Laura Stapleton Introduction Recent years have seen remarkable development in estimating average treatment effects
More informationEC352 Econometric Methods: Week 07
EC352 Econometric Methods: Week 07 Gordon Kemp Department of Economics, University of Essex 1 / 25 Outline Panel Data (continued) Random Eects Estimation and Clustering Dynamic Models Validity & Threats
More informationRecent advances in non-experimental comparison group designs
Recent advances in non-experimental comparison group designs Elizabeth Stuart Johns Hopkins Bloomberg School of Public Health Department of Mental Health Department of Biostatistics Department of Health
More informationEconometric Game 2012: infants birthweight?
Econometric Game 2012: How does maternal smoking during pregnancy affect infants birthweight? Case A April 18, 2012 1 Introduction Low birthweight is associated with adverse health related and economic
More informationComparisons of Dynamic Treatment Regimes using Observational Data
Comparisons of Dynamic Treatment Regimes using Observational Data Bryan Blette University of North Carolina at Chapel Hill 4/19/18 Blette (UNC) BIOS 740 Final Presentation 4/19/18 1 / 15 Overview 1 Motivation
More informationCOMMITTEE FOR PROPRIETARY MEDICINAL PRODUCTS (CPMP) POINTS TO CONSIDER ON MISSING DATA
The European Agency for the Evaluation of Medicinal Products Evaluation of Medicines for Human Use London, 15 November 2001 CPMP/EWP/1776/99 COMMITTEE FOR PROPRIETARY MEDICINAL PRODUCTS (CPMP) POINTS TO
More informationSawtooth Software. The Number of Levels Effect in Conjoint: Where Does It Come From and Can It Be Eliminated? RESEARCH PAPER SERIES
Sawtooth Software RESEARCH PAPER SERIES The Number of Levels Effect in Conjoint: Where Does It Come From and Can It Be Eliminated? Dick Wittink, Yale University Joel Huber, Duke University Peter Zandan,
More information11/24/2017. Do not imply a cause-and-effect relationship
Correlational research is used to describe the relationship between two or more naturally occurring variables. Is age related to political conservativism? Are highly extraverted people less afraid of rejection
More information